Dynamic Subscriber Network Physical Impairment Detection Techniques

1. A method comprising: providing input to multiple estimator models that are each trained to output a conditional likelihood with respect to a particular feature subset ensemble from among different feature subset ensembles; obtaining, by a computing device, data for a line of a subscriber network that indicates a set of one or more scores, wherein each score included in the set of scores indicates the conditional likelihood that the line has a type of impairment with respect to a different feature subset ensemble; providing, by the computing device, the obtained data as input to a prediction model that is trained to output, for each of different sets of feature subset ensembles, a confidence score representing an overall likelihood that a particular line has a physical impairment; receiving, by the computing device and from the prediction model, data indicating a particular confidence score representing an overall likelihood that the line has the physical impairment; and providing, by the computing device, the particular confidence score for output.

2. The method of claim 1 , wherein each feature subset ensemble specifies a type of network measurement that reflects a particular set of signal attributes.

3. The method of claim 2 , the prediction model is a perceptron that is trained to (i) identify a respective weight assigned to each of the network measurements specified in the feature subset ensembles, and (ii) compute the confidence score by combining the scores according to the identified weights.

4. The method of claim 1 , wherein the multiple estimator models include a signal to noise ratio (SNR) model, and a quiet line noise (QLN) model.

5. The method of claim 1 , wherein the subscriber network is a digital subscriber line (DSL) network.

6. The method of claim 1 , wherein the subscriber network is a passive optical network (PON).

7. A method, comprising: obtaining, by a computing device, data for a line of a subscriber network that indicates a set of one or more scores, wherein each score included in the set of scores indicates a conditional likelihood that the line has a type of impairment with respect to different feature subset ensembles; providing, by the computing device, the obtained data as input to a model that is trained to output, for each of different sets of feature subset ensembles, a confidence score representing an overall likelihood that a particular line has a physical impairment; receiving, by the computing device and from the model, data indicating a particular confidence score representing an overall likelihood that the line has the physical impairment; determining, by the computing device, a location along the line of the subscriber network that is associated with the physical impairment; and providing, by the computing device, a report indicating the location for output to a service provider system associated with the subscriber network.

8. A method, comprising: obtaining, by a computing device, data for a line of a subscriber network that indicates a set of one or more scores, wherein each score included in the set of scores indicates a conditional likelihood that the line has a type of impairment with respect to different feature subset ensembles; providing, by the computing device, the obtained data as input to a model that is trained to output, for each of different sets of feature subset ensembles, a confidence score representing an overall likelihood that a particular line has a physical impairment; receiving, by the computing device and from the model, data indicating a particular confidence score representing an overall likelihood that the line has the physical impairment; determining, by the computing device, that the particular confidence score satisfies a predetermined threshold; and based on determining that the particular confidence score satisfies the predetermined threshold, providing, by the computing device, a report to a service provider system associated with the subscriber network, the report indicating that the line of the subscriber network is predicted to have the physical impairment.

9. A method, comprising: obtaining, by a computing device, data for a line of a subscriber network that indicates a set of one or more scores, wherein each score included in the set of scores indicates a conditional likelihood that the line has a type of impairment with respect to different feature subset ensembles; providing, by the computing device, the obtained data as input to a model that is trained to output, for each of different sets of feature subset ensembles, a confidence score representing an overall likelihood that a particular line has a physical impairment; receiving, by the computing device and from the model, data indicating a particular confidence score representing an overall likelihood that the line has the physical impairment; and providing, by the computing device, the particular confidence score for output, wherein: the obtained data for a line of a subscriber network indicates activity on the line of the subscriber network during a particular time period; and the particular confidence score indicates an overall likelihood that the line has the physical impairment during the particular time period.

10. A system comprising: one or more computers; and one or more storage devices storing instructions that, when executed by the one or more computers, cause the one or more computers to perform operations comprising: providing input to multiple estimator models that are each trained to output a conditional likelihood with respect to a particular feature subset ensemble from among different feature subset ensembles; obtaining, by a computing device, data for a line of a subscriber network that indicates a set of one or more scores, wherein each score included in the set of scores indicates the conditional likelihood that the line has a type of impairment with respect to a different feature subset ensemble; providing, by the computing device, the obtained data as input to a prediction model that is trained to output, for each of different sets of feature subset ensembles, a confidence score representing an overall likelihood that a particular line has a physical impairment; receiving, by the computing device and from the prediction model, data indicating a particular confidence score representing an overall likelihood that the line has the physical impairment; and providing, by the computing device, the particular confidence score for output.

11. The system of claim 10 , wherein each feature subset ensemble specifies a type of network measurement that reflects a particular set of signal attributes.

12. The system of claim 11 , the prediction model is a perceptron that is trained to (i) identify a respective weight assigned to each of the network measurements specified in the feature subset ensembles, and (ii) compute the confidence score by combining the scores according to the identified weights.

13. A non-transitory computer-readable storage device encoded with computer program instructions that, when executed by one or more computers, cause the one or more computers to perform operations comprising: providing input to multiple estimator models that are each trained to output a conditional likelihood with respect to a particular feature subset ensemble from among different feature subset ensembles; obtaining, by a computing device, data for a line of a subscriber network that indicates a set of one or more scores, wherein each score included in the set of scores indicates the conditional likelihood that the line has a type of impairment with respect to a different feature subset ensemble; providing, by the computing device, the obtained data as input to a prediction model that is trained to output, for each of different sets of feature subset ensembles, a confidence score representing an overall likelihood that a particular line has a physical impairment; receiving, by the computing device and from the prediction model, data indicating a particular confidence score representing an overall likelihood that the line has the physical impairment; and providing, by the computing device, the particular confidence score for output.

14. The device of claim 13 , wherein each feature subset ensemble specifies a type of network measurement that reflects a particular set of signal attributes.

15. The device of claim 14 , the prediction model is a perceptron that is trained to (i) identify a respective weight assigned to each of the network measurements specified in the feature subset ensembles, and (ii) compute the confidence score by combining the scores according to the identified weights.

16. The device of claim 13 , wherein the multiple estimator models include a signal to noise ratio (SNR) model, and a quiet line noise (QLN) model.

17. The device of claim 13 , wherein the subscriber network is a digital subscriber line (DSL) network.